id stringclasses 9
values | family stringclasses 9
values | tier stringclasses 3
values | instruction stringclasses 9
values | input stringclasses 9
values | answer unknown | verify stringclasses 5
values | provenance unknown | license stringclasses 2
values |
|---|---|---|---|---|---|---|---|---|
brv-gl-03 | crossarch_equivalence | medium | BLOCK-A and BLOCK-B below are scrubbed disassembly of two single functions from production Debian/Ubuntu glibc 2.39 cross-toolchain builds (AArch64, RISC-V, MIPS64r6 n64, PowerPC64 little-endian). Each block header names its ISA. Branch-target symbol labels and numeric relocation comments were removed; MIPS blocks keep... | BLOCK-A (AArch64):
9bfc4: sub x10, x1, x0
9bfc8: mov x8, #0x101010101010101
9bfcc: and x6, x0, #0x7
9bfd0: tst x10, #0x7
9bfd4: b.ne 9c054
9bfd8: cbnz x6, 9c030
9bfdc: nop
9bfe0: ldr x3, [x0, x10]
9bfe4: ldr x2, [x0], #8
9bfe8: sub x4, x2, x8
9bfec: orr x6, x2, #0x7f7f7f7f7f7f7f7f
9bff0: bics x4, x4, x6
9bff4: ccmp x2,... | {
"same": true,
"symbol": "strcmp"
} | detailed rubric | {
"binary": "/usr/aarch64-linux-gnu/lib/libc.so.6 ; /usr/mips64el-linux-gnuabi64/lib/libc.so.6",
"sha256_16": "101106b8ae666012 ; 29e42020c092abb1",
"addr": "0x9bfc4 ; 0xd8270",
"arch": "AArch64 ; MIPS R3000"
} | (c) Samuel Anttila / RE-Lab - commercial license on purchase |
brv-ag-10 | agentic_flag_hunt | hard | You have the binary hightwist in your working directory (x86-64 static ELF, symbols stripped). Find the exact argv[1] that makes it print OK. Use any static-analysis tools available; you may also execute it. | binary file: hightwist | "Vyrtane_2cK9" | exact argv match; program prints OK | {
"binary": "bins/hightwist",
"sha256_16": "24ea8b38101e457e",
"arch": "x86-64"
} | (c) Samuel Anttila / RE-Lab - commercial license on purchase |
brv-gl-19 | func_summary | medium | The block below is scrubbed disassembly of one function from AArch64 glibc 2.39 (libc.so.6) at 0x9dd60. Symbol labels and numeric comments removed. Summarize precisely what this function does and name it: give the exact exported symbol, or for an IFUNC resolver the symbol whose implementation it selects. | 9dd60: adrp x2, 1af000
9dd64: ldr x2, [x2, #3728]
9dd68: stp x29, x30, [sp, #-16]!
9dd6c: adrp x1, a2000
9dd70: mov x29, sp
9dd74: ldr x2, [x2, #552]
9dd78: add x1, x1, #0xd40
9dd7c: adrp x0, a2000
9dd80: add x0, x0, #0xc00
9dd84: ldp x29, x30, [sp], #16
9dd88: tst x2, #0x40000
9dd8c: csel x0, x0, x1, eq
9dd90: ret | "IFUNC resolver for strlen (exported symbol strlen is an STT_GNU_IFUNC; this runs once at load time, it contains no string code). It loads GLRO(dl_hwcap) from the loader's _rtld_global_ro block (adrp + ldr [x2,#3728], then ldr [x2,#552]), tests bit 0x40000 = HWCAP_ASIMD (1<<14) with tst, and returns via csel one of two... | detailed rubric | {
"binary": "/usr/aarch64-linux-gnu/lib/libc.so.6",
"sha256_16": "101106b8ae666012",
"addr": "0x9dd60",
"arch": "AArch64"
} | (c) Samuel Anttila / RE-Lab - commercial license on purchase |
brv-relf-012 | func_id_exact | hard | This is Ghidra x86-64 disassembly (excerpt) of one function from libz.so.1. Identify the exact symbol name of this function. | 0x106be6: LEA RAX, [RCX + 0x15b0]
...
0x106c20: MOVZX EAX, byte ptr [R8] ; 16-byte unrolled sum chain
0x106c37: ADD RAX, R15
0x106c44: ADD R14, RAX
0x106c51: ADD R13, R14
0x106c61: ADD R12, R13
... (running s2 as sum of running s1 over 16 bytes) ...
0x106cf9: JNZ 0x00106c20
0x106cff: MOV RAX, 0xf00e10d2fc5cd ... | "adler32_z" | exact symbol match after normalization (case-insensitive, ignore version suffixes). | {
"binary": "bins/libz.so.1",
"sha256_16": "f61677c7bdb7c2aa",
"addr": "0x6ba0",
"size": 1716
} | (c) Samuel Anttila / RE-Lab - commercial license on purchase |
funcidmc-a6d1ab7099b62132-4f70 | func_id_mc | medium | Identify the function from its disassembly. Reply with the correct option. | 4f70: f3 0f 1e fa endbr64
4f74: 55 push %rbp
4f75: 48 89 e5 mov %rsp,%rbp
4f78: 53 push %rbx
4f79: 48 81 ec 00 10 00 00 sub $0x1000,%rsp
4f80: 48 83 0c 24 00 orq $0x0,(%rsp)
4f85: 48 81 ec e8 00 00 00 sub $... | {
"correct": "lutil_debug",
"options": [
"ber_bvreplace_x",
"ber_get_bitstringa",
"ber_strndup",
"lutil_debug"
]
} | exact option match | {
"binary": "liblber.so.2.0.200",
"sha256_16": "a6d1ab7099b62132",
"addr": "0x4f70",
"size": 265
} | (c) RE-Lab — commercial license on purchase |
brv-xarch-011 | arch_semantics_qa | medium | AArch64: x0 is a 64-bit value and w1 an unsigned count. What does this three-instruction function compute? Derivable from the shown instructions only. | neg w1, w1
ror x0, x0, x1
ret | "It rotates x0 left by the amount in w1. AArch64 ror normalizes its shift amount modulo the register width (64), so rotating right by (-r mod 64) equals rotating left by r; neg first forms the two's-complement of the count (and zeroes the upper half of x1, which is irrelevant since only the low 6 bits steer ror). Net: ... | detailed rubric | {
"binary": "bins/xhash-aarch64",
"sha256_16": "02c89b8c70ab496a",
"addr": "0x400834",
"arch": "AArch64"
} | (c) Samuel Anttila / RE-Lab - commercial license on purchase |
brv-isa-14 | isa_semantics_qa | medium | RISC-V fragment; on entry a0 holds the argument. State what value ends up in a0 at ret, derived from the shown instructions only. | mv a5,a0
li a0,0
beqz a5,L1
L0:
addiw a4,a5,-1
and a5,a5,a4
sext.w a5,a5
addiw a0,a0,1
bnez a5,L0
ret
L1:
ret | "a0 = number of set bits (population count) of the low 32 bits of the original a0. mv copies the argument into a5 and li zeroes the counter a0; the loop body addiw -1 / and implements x &= x-1 (clears the lowest set bit), sext.w re-sign-extends to keep beqz/bnez correct on 32-bit values, and addiw +1 increments the cou... | detailed rubric | {
"binary": "bins/xrot-riscv64",
"sha256_16": "ccd3e8a49afbbb6d",
"addr": "0x105b8",
"arch": "RISC-V"
} | (c) Samuel Anttila / RE-Lab - commercial license on purchase |
brv-repe-013 | pe_structure_qa | easy | From this PE header excerpt, answer: (1) 32-bit or 64-bit and how you know, (2) EXE or DLL and which field says the intended environment, (3) which linker/toolchain family produced it. | File header: Machine=0x8664, NumberOfSections=6. Optional header: Magic=0x20B, MajorLinkerVersion=14, MinorLinkerVersion=51, AddressOfEntryPoint RVA=0x1820, ImageBase=0x140000000, SizeOfImage=0x28000, Subsystem=3. | "(1) 64-bit: Magic 0x20B is PE32+ and Machine 0x8664 is IMAGE_FILE_MACHINE_AMD64. (2) Console-subsystem image: Subsystem=3 is IMAGE_SUBSYSTEM_WINDOWS_CUI, i.e. a text-mode program (console executables are EXEs; the preferred ImageBase 0x140000000 is the MSVC x64 default). (3) Linker version 14.51 belongs to Microsoft's... | exact field match | {
"binary": "bins/decoder-x64.exe",
"sha256_16": "14870d2ab2c9a5fb",
"addr": "0x140000118"
} | (c) Samuel Anttila / RE-Lab - commercial license on purchase |
brv-repe-003 | pe_import_reasoning | medium | Given this Windows x64 executable's import table and application strings, what is the program most likely doing? Justify from imports/strings; flag what the evidence does not support. | Imports (all from KERNEL32.dll): GetCommandLineA/W, GetStdHandle, WriteConsoleW, WriteFile, ExitProcess, HeapAlloc/HeapFree/VirtualProtect, LoadLibraryExW/GetProcAddress, standard CRT startup/unwind/locale helpers. No other DLL is imported. Selected strings: "usage: vmcheck <input>", "accepted", "rejected". | "A one-argument accept/reject gate: `prog <input>` prints either \"accepted\" or \"rejected\"; with no argument it prints the usage line. The binary vocabulary accepted/rejected plus a single <input> parameter indicates a boolean validator over the argument string, decided entirely in-process: the KERNEL32-only import ... | detailed rubric | {
"binary": "bins/vmcheck-x64.exe",
"sha256_16": "24edb05e228ecd29",
"addr": "0x140010380"
} | (c) Samuel Anttila / RE-Lab - commercial license on purchase |
binary-reasoning-v1
Verified reverse-engineering tasks over real production binaries — an RL / eval environment pack where every answer is derived from the artifact itself and machine-checked.
This public set is an entry-level demo and is not the current product. See re-lab.dev. It illustrates the task format and grading, and frontier models solve it easily.
- The full pack, not currently offered: 10,251 items · 10 families · 7 ISAs (x86-64, i686, AArch64, ARM32, RISC-V, MIPS o32/n64, PPC64LE)
- Sources: 649 distinct third-party file names referenced in shipped provenance (system libs incl. libcrypto/zstd/sqlite3/libxml2, Mesa DRI drivers, QEMU device models, systemd internals), Debian production cross-toolchain glibc, Android/Termux production aarch64 libs, curated Windows PEs, labeled cross-compiles
- Provenance per item: binary name and sha256_16 on every item; function address/size and architecture where applicable
- Verification: automated re-verification through our in-house analysis pipeline (answers re-verified against the binaries); 10,172
deterministic items score 1.0 on gold / 0.0 on abstention; automated answer-leak scan found 0/10,251 hits; all tasks load
against the real
verifiersv1 framework; 29 agentic challenges ship Harbor-format task packages, built + scored in real Docker containers with reward-chain gold and wrong-answer tests passing; each binary was also checked with 24 single-character mutations that scored zero - Baseline: a small local open-weight model mini-eval scored
5/23(22%) on selected static deterministic items; seeEVAL_RESULTS.mdfor conditions. Frontier reference: Claude Opus 5.5 with shell and file-read tools solved all 29 public agentic environments (145/145 attempts, 2026-10-29; non-isolated upper bound, see BENCHMARK-REPORT.md), so the public tier is warm-up level.
What a harder private set measured
One sealed run on 6 Nov 2026, through the provider's API. Eight tasks built from our own source were held out of development. The rules and budgets were fixed and timestamped before the first start, and the verdicts stayed unread until the run had closed.
| Model | Tasks tried | Passed on the first attempt | Passed if one shared integer-parsing rule is not counted |
|---|---|---|---|
| Claude Haiku 5.5 | 8 | 1 | 2 |
| Claude Sonnet 5.5 | 8 | 7 | 8 |
| Claude Opus 5.5 | 3 | 1 | 2 |
- What it shows: this set separates models below Sonnet from Sonnet.
- What it does not show: it is not frontier-hard, and it must not be used to rank top models. Sonnet passes nearly all of it. Three attempts cannot place Opus: on the three tasks both ran, Sonnet passed 3 and Opus 1. Both Opus failures missed a single test case (1 of 67 and 1 of 94), and Opus stopped sooner, with a median of 12.6 minutes against 21.2.
- How far to trust it: these are small numbers. The exact interval for Sonnet's 7 of 8 runs from 47% to 99.7%, and real uncertainty may be wider, because the tasks come from one generator.
- Conduct of the run: 44 starts, none refused by the provider. One start was lost to a network drop on our side and re-run once, as the rules allowed.
- Receipts: entry
be79ce047b6de83ea2a5f462634c398df0d2121c763bffd8db4191d2c50c8e86, closing38dad74ffe572a615af4a62c17226cc484012f548abc9c0e02c662f6bc339b19, both timestamped by DigiCert and FreeTSA. The tasks stay private.
Contents of this repo (free sample)
| file | what |
|---|---|
DATASHEET.md |
full composition table, guarantees, known limits, positioning vs existing work |
sample_free.jsonl |
9 items — one per family |
EVAL_RESULTS.md |
baseline eval methodology + results (incl. a frontier reference: Opus 5.5 single-shot ~58% on ID items) |
EVAL_RESULTS-passk.md |
full 29-env agentic pass@5 result for Claude Opus 5.5 with conditions and caveats |
BENCHMARK-REPORT.md |
benchmark report for the public agentic environments, including intervals, method notes, and known limits |
VERIFICATION-DOSSIER-PUBLIC.md |
public proof summary with measured checks and known caveats |
verification_kit/ |
sample-level grader, wrong-answer, key-sensitivity, and leak-check scripts; see verification_kit/README.md for limitations |
environment/src/binary_reasoning_v1/grading.py |
minimal grader module used by the verification kit. The HF free sample here is 9 items; the Prime Intellect environment package exposes a separate 100-item open sample. |
harbor_pilot_29.zip |
all 29 agentic Harbor task packages (task.toml + instruction + Dockerfile + verifier + solution scripts and solution explanations), ready for harbor run or a free HUD import (direct download) |
Also live on the Prime Intellect Environments Hub: s3rl4/binary-reasoning-v1
(prime env install s3rl4/binary-reasoning-v1).
The full pack (10,251 items, the gated companion repo S3RL4/binary-reasoning-v1-full) is not currently offered.
Custom task families, built from our own source on written commission: samueljanttila@gmail.com.
Privacy: PRIVACY.md
License
Task annotations, graders and packaging © RE-Lab, all rights reserved pending a commercial licence; third-party binaries remain under their own licences (provenance per item). Research evaluation use is granted for the free sample.
- Downloads last month
- 679